Re: control charts and non-normality

[DANSWART@aol.com]

In a message dated 4/25/00 tpr@po.cwru.edu (Tom Ryan) wrote:

<< Appeals have been made to Tchebysheff's Inequality, but control chart 
properties depend on the *reciprocal* of 
the percentage of observations that can be expected to fall outside the 
control limits, and not on the percentage itself. >>

In a message dated 05/08/2000 2:57:25 PM Pacific Daylight Time, 
GrantBlair@aol.com writes:

<< Quoting from "SPC for Practitioners: Special Causes and Continuous 
Processes" by Gary Fellers "As a rule of thumb, there should be an average of 
20 plotted points between occurences of assignable cause. After approximately 
100 plottings with SPC in effect, the charts can tell you if the sampling 
interval is correct, and the process is in a state of statistical control" >>


Assuming out of control events display predictable patterns on control 
charts...  Whenever we speak of "expected values" don't we run the 
significant risk of imposing a pattern (or lack of pattern) to otherwise 
random observations, and thus risk tampering with the control chart process 
itself?  There can be a surprising number of consecutive runs of heads or 
tails from a fair coin.  For instance, when Gary Fellers speaks of "an 
AVERAGE of 20 points between occurrences of assignable cause" I read this to 
mean that a very large number of control charts over a very long period of 
time can expect to display such an average (a prediction based on enumerative 
methods).  But I don't think he means to predict any such pattern in a 
specific group of points, like say the next 50 or 100 or 350 observations 
does he?   Is his prediction is based on the assumptions of enumerative 
statistical methods?  It seems to me that enumerative statistics is very good 
at describing groups of things taken as a whole, but is not good at all at 
specific, individual predictions.  Does that apply here?

(I will be interested in following the debates as to assigning probabilities 
to patterns on control charts)

Thank you for reading.

Dan
danswart@aol.com
=======================================================================